Paper

APS-VLA: Action-Guided and Future-Aware Token Pruning for Efficient Vision-Language-Action Models · ByteTrail